doc.: IEEE /227 Submission May 2001 S. Gray, NokiaSlide 1 Throughput and Loss Packet Performance of DCF with Variable Transmit Power Steven D. Gray and Venkatesh Vadde [steven.gray, Nokia Research Center Irving, TX

doc.: IEEE /227 Submission May 2001 S. Gray, NokiaSlide 2 Outline Motivation CCA performance –Probability of detection analytical model –Simulation Test cases DCF simulation –Throughput and delay

doc.: IEEE /227 Submission May 2001 S. Gray, NokiaSlide 3 Motivation for Analysis Proposals have been offered to allow STAs to adjust power on an individual basis –In discussion, many expressed the opinion that this will cause “hidden terminals” –Quantitative results are offered to illustrate the performance (delay and throughput) of a networking using two different power settings (E)DCF and PCF/HCF –DCF STAs must transmit at a power level to achieve a target packet loss rate AND keep others off the channel –PCF/HCF must only transmit at a power level to achieve a target packet loss rate

doc.: IEEE /227 Submission May 2001 S. Gray, NokiaSlide 4 CCA Probability of Detection IEEE802.11a states: “A start of a valid OFDM transmission at receiver level equal or greater than 6 Mbits/s sensitivity (-82dBm) shall cause CCA to indicate Busy with probability >90% within 4 us. If the preamble portion was missed, the receiver shall hold the CS signal Busy for any signal 20 dB above the minimum 6 Mbits/s sensitivity (-62 dBm)”

doc.: IEEE /227 Submission May 2001 S. Gray, NokiaSlide 5 CCA Test: Analytical Model Many ways exist for doing CCA. One method is to use a delay and correlate filter based upon the periodicity of the PLCP Preamble General structure of a delay and correlate filter are the sampled time-series from the PLCP Preamble D the periodicity in the short symbols

doc.: IEEE /227 Submission May 2001 S. Gray, NokiaSlide 6 CCA Test: Analytical Model General approach for CCA is to define where is a normalizing signal such that

doc.: IEEE /227 Submission May 2001 S. Gray, NokiaSlide 7 CCA Test: Analytical Model Consider a specular channel Let the received signal be modeled as is the transmitted OFDM signal is background noise

doc.: IEEE /227 Submission May 2001 S. Gray, NokiaSlide 8 CCA Test: Analytical Model Ifis complex Gaussian then is exponential The essence of a delay and correlate filter is where conditioned on a valid PLCP preamble being sent

doc.: IEEE /227 Submission May 2001 S. Gray, NokiaSlide 9 CCA Test: Analytical Model Probability of detection can be defined by considering the distribution of a sum of non iid exponential random variables (CLT can not typically be invoked because number of multipaths are few in number i.e., five) For ease of analysis, two multipaths are considered

doc.: IEEE /227 Submission May 2001 S. Gray, NokiaSlide 10 CCA Test: Analytical Model Ifis distributed exponential with then if the pdf for z is

doc.: IEEE /227 Submission May 2001 S. Gray, NokiaSlide 11 CCA Test: Analytical Model To simplify analysis, background noise is modeled as a constant level The probability of detection can be defined as

doc.: IEEE /227 Submission May 2001 S. Gray, NokiaSlide 12 Test Case PLCP Preamble Detection AP STA 1 STA 2 60 meters OBS 63 meters OBS 20 meters LOS SNR STA1 = 8 dBSNR STA2 = 8 dB TX Power = 5 dBm TX Power = 23 dBm The signal power of STA 1 seen at STA2 = -11 dB The signal power of STA 2 seen at STA1 = 7 dB

doc.: IEEE /227 Submission May 2001 S. Gray, NokiaSlide 13 Test Case Line of Sight (LOS) path loss Obstucted Line of Sight (OBS) path loss Transmitter Antenna Height LOS n 1 n 2 OBS n Low (3.7 m) Medium (8.5 m) High (13.3 m) Fresnel distance Reference path loss (1 meter) Feuerstein, M.J., Blackard, K.L., Rappaport, T.S., Seidel, S.Y., and Xia, H.H., “Path Loss, Delay Spread and Outage Models as Functions of Antenna Height for Microcellular System Design,” IEEE Transactions on Vehicular Technology, Vol. 43, No 3, pp August 1994.

doc.: IEEE /227 Submission May 2001 S. Gray, NokiaSlide 14 Test Case Two tap channel  tap 1 = 0.7  tap 2 = 0.3 Probability of false alarm at SNR = 6 dB is approx 9 % Pd = at SNR = - 11dB

doc.: IEEE /227 Submission May 2001 S. Gray, NokiaSlide 15 DCF Simulation Procedure Traffic evenly distributed between different users Poisson arrivals for traffic; Uniformly distributed packet sizes [500, 1500] bits 12Mb/s PHY assumed, average packet length = 1000 bits Average PHY-related packet error rate = 1% Simulations averaged over 10 6 slot-times

doc.: IEEE /227 Submission May 2001 S. Gray, NokiaSlide 16 Hidden-node DCF Simulations Assume users, randomly allocated power levels 2 power-levels defined: High and Low Probability{detection of low-power users} = 1% Probability{detection of high-power users} = 90% Number of retransmissions allowed = 3 Initial back-off value = 16 slot-times (1 slot-time = 9 micro-sec)

doc.: IEEE /227 Submission May 2001 S. Gray, NokiaSlide 17 Throughput Results % of network throughput can be lost with hidden-nodes

doc.: IEEE /227 Submission May 2001 S. Gray, NokiaSlide 18 Throughput Results - 2

doc.: IEEE /227 Submission May 2001 S. Gray, NokiaSlide 19 Throughput of High-vs-Low Power Packets Network composition assumed = 50%low:50% High Conclusion: Low power STAs suffer poor throughput as load is increased

doc.: IEEE /227 Submission May 2001 S. Gray, NokiaSlide 20 Dropped Packet Results Up to 70 % increase in dropped packets for network

doc.: IEEE /227 Submission May 2001 S. Gray, NokiaSlide 21 Conclusions Throughput is decreased when large variations of power are used by STAs in a BSS Low power STAs increase the number of hidden terminals –Multiple retransmission of packets required to ensure success –Battery saving desired with power reduction may be lost